A Reinforcement Learning Approach for Modeling Organic Compound-Induced Antimicrobial Resistance Dynamics

This study investigates the use of reinforcement learning (RL) to model antimicrobial resistance (AMR) dynamics driven by copper exposure. In a simulated environment, three agent strategies were evaluated: (1) random, (2) rule-based, and (3) Q-learning. These agents were compared to assess how varying levels of adaptability affect bacterial resistance to chloramphenicol and polymyxin B. The Q-learning agent consistently achieved lower resistance levels and copper burden compared to both the rule-based and random agents, demonstrating its potential for optimizing AMR management. The rule-based agent showed intermediate performance, while the random agent exhibited higher variability in resistance trajectories. This study introduces an exploratory RL framework for simulating AMR control strategies, especially in scenarios where traditional diagnostic tools are unavailable. The results demonstrate that RL can provide valuable insights into optimizing AMR management, particularly in settings with limited access to traditional diagnostic tools and resources.